CUSTOMIZED SPEECH AND THE FIRST AMENDMENT.

• Author: Rauch, Daniel E.

TABLE OF CONTENTS I. INTRODUCTION 407 II. THE LAW OF CUSTOMIZED SPEECH 413 A. Speaker Use of Audience Information 414 1. Plenary Protection for Content and Audience Choices 415 2. Uncapped Persuasive Efficacy 416 3. Speaker Autonomy 422 4. Audience-Information Uses and the First Amendment 424 B. Speaker Collection of Audience Information 425 1. No Right to Government-Created Audience Information 426 2. Generally Applicable Collection Laws Permitted 426 3. Truly Essential Speech Inputs Protected 427 C. The Customized Speech Doctrine 428 D. General Objections 430 III. DIGITAL CUSTOMIZED SPEECH 433 A. Deeper Data Collection 433 B. Stronger Targeting and Tailoring 435 C. Advertising Megaplatforms 437 D. Few Collection Limits 438 IV. CUSTOMIZED SPEECH AND DEMOCRACY 439 A. Democratic Discourse Harms 440 B. Partisanship Harms 443 C. Autonomy Harms 445 1. Nonrational Appeals 446 2. Exploiting Personal Frailties and Vulnerabilities 447 3. Strong-Form Mind Control 450 D. Countervailing Benefits 452 1. Informing and Engaging the Electorate 452 2. Empowering the Marginalized 454 3. Checking Government Overreach 456 V. ASSESSING CUSTOMIZED SPEECH RESTRICTIONS 458 A. Direct Limits on Use of Audience Information 459 B. Customized Speech Disclosure Requirements 460 C. Generally Applicable Information Collection Laws 463 D. Voluntary Political Action 464 VI. CONCLUSION 465 I. INTRODUCTION

A civil rights leader only shares bus boycott flyers with groups she knows support racial justice. (1) A polyglot politician gives four different speeches--in four different languages--to his Yiddish-, English-, Italian-, and Croatian-speaking constituents. (2) A conservative candidate uses mailing lists from ideological allies to choose where to send campaign literature. (3) A gay rights coalition uses opinion polls to craft messages that resonate with swing voters. (4) A consultancy scrapes data from fifty million Facebook profiles, then creates and sends personalized ads based on each target's "psychographic" profile. (5)

Each of these practices is Customized Speech: speech targeted or tailored based on knowledge of one's audience. The first four are familiar and uncontroversial--politics as usual. The fifth, practiced by Cambridge Analytica in the 2016 U.S. presidential election, shook the world. This Article explains why, doctrinally, the first four of these customizations enjoy robust First Amendment protection. And it explains why this doctrinal conclusion is normatively sound, even when applied to the Digital Customized Speech at the heart of the final tactic.

Customized Speech is pervasive, permeating our rhetorical traditions, (6) our relationships, (7) and perhaps most prominently, our politics. (8) But despite (or perhaps because of) this ubiquity, courts and commentators have yet to offer any coherent theory of Customized Speech and the First Amendment. Some assert, in brief asides, that restrictions on Speech Customization do not "limit[] the manner in which [one] can speak," and so pose little constitutional challenge. (9) Others, with equal brevity, suggest the opposite: That the use of audience information to target or tailor speech must surely enjoy strong protection. (10) Largely, though, judges and scholars have said nothing at all, leaving the doctrinal question of Customized Speech unanswered. (11)

Since 2016, the salience of this gap has grown dramatically. That year, Cambridge Analytica, a British consultancy, undertook a massive project of political Speech Customization. It harvested Facebook data from millions of American users, then analyzed the data to create and distribute "psychographic" advertising--ads ostensibly targeted to each user's emotional and psychological attributes. (12) And, the story goes, it succeeded, swinging just enough votes to put Donald Trump in the White House. (13)

Factually, this tale is largely inaccurate, greatly overstating the consultancy's competence and impact. (14) But narratively, it struck a chord. For despite its brazen crimes, (15) Cambridge Analytica's core model was a set of Speech Customization techniques now common in American politics: pooling vast seas of voters' personal information, analyzing this data to guess which voters will embrace which message, running thousands of tests to see which guesses are correct, then sending the most persuasive ad, at the most persuasive time, to the most persuadable voter. (16)

Galvanized by the Cambridge Analytica scandal, a near-consensus of scholars, commentators, and legislators has concluded that machine learning, social networks, and Big Data make modern, Digital Customized Speech a threat to our democracy. Privacy and election law scholars argue such practices impoverish democratic discourse, (17) stoke partisan hatred, (18) and shackle personal autonomy. (19) Commentators from tech titans to top regulators warn that targeted and tailored politics pose grievous societal harms. (20) Heeding such calls, lawmakers have proposed (and enacted) unprecedented limits on political Digital Customized Speech, from audience-targeting disclosure rules (21) to outright curbs and bans. (22)

Yet despite this flurry of activity, we still lack a coherent understanding of how Speech Customization interacts with the First Amendment. Indeed, on the few occasions proponents of Customized Speech limits acknowledge First Amendment "hurdle[s]," (23) they do so only in passing, offering little insight as to just what sort of hurdles are at stake (other than to assert that their preferred reforms will clear them). (24) Thus, even as courts confront the first wave of challenges to Digital Customized Speech restrictions, (25) they, along with legislators, litigators, and technologists, are without a doctrinal roadmap.

This Article provides that map. In doing so, it makes three novel, timely, and significant contributions. First, it provides a rigorous doctrinal analysis as to whether, and to what extent, the First Amendment protects political Customized Speech. (26) Here, I find a speaker's use of lawfully obtained information about her audience when customizing political speech is, itself, core political speech. This conclusion flows from a speaker's nearly plenary right to choose political speech's content and audience; her correlative right to use the information, knowledge, and other resources she possesses to make such choices as effectively as she can (and to make them in accordance with her inherent autonomy); and the special role of audience information as just such a vital persuasive resource.

Turning from a speaker's use of audience information to her initial collection of such information, I find that collection conduct--while not entitled to the high protections of audience-information use--still implicates First Amendment values. Here, I distill three principles: First, the state has no constitutional obligation to provide speakers with government-created audience information (like voter records or census data), even if such information would be very helpful to those speakers. Second, the state may, by generally applicable conduct laws, limit audience-information collection practices (hence, general laws against trespass apply even to journalists). But because of the First Amendment valence of audience-information collection, even such generally applicable laws must be carefully drawn to avoid discriminating based on speaker content or viewpoint. And third, even otherwise acceptable collection limits may still fail if they unduly burden inputs truly essential to core First Amendment activity (like blanket bans on videotaping or onerous taxes on newsprint).

Combined, these principles offer a clear doctrinal structure for Speech Customization: The state may regulate audience-information collection, provided it does so through laws that are content-neutral, generally applicable, carefully drawn, and not unduly burdensome to essential speech inputs. But whether or not the state enacts such collection limits, it remains almost powerless to proscribe a speaker's use of otherwise lawfully collected audience information for political Customized Speech.

After establishing this doctrinal framework, I next consider Digital Customized Speech: the Customized Speech enabled by tools like Big Data analytics, massive social networks, and bleeding-edge machine learning. Here, I outline four trends driving this communicative revolution: vast aggregations of digital personal data (from financial transactions to personality insights); staggering tools to target, tailor, and track the effectiveness of messaging; Advertising Megaplatforms (like Google and Facebook) that "rent out" prodigious customization capabilities to speakers of all sizes and persuasions; and a legal environment with few meaningful limits on digital personal data collection.

Responding to these dynamics and the ways they have played out in recent elections, scholars, commentators, and legislators increasingly see Digital Customized Speech as a threat to democratic values, one that constitutionally can, and normatively should, be curtailed. As its second major contribution, this Article takes on, and rejects, this near-consensus view. (27) While Digital Customized Speech raises concerns about democracy, partisanship, and autonomy, these are less severe than critics claim. Further, such harms must be weighed against countervailing benefits. Protection for Digital Customized Speech helps inform and engage the electorate, empowers marginalized groups, and is a valuable check on government overreach. Thus, the harms attributed to Digital Customized Speech, as of now, are not the sort of immediate, unambiguous, and overwhelming dangers that might justify abandoning longstanding doctrinal principles.

Finally, as a third significant contribution, I apply these insights to evaluate four proposed approaches for regulating Digital Customized Speech: direct limits on speakers' use of audience...